File size: 5,210 Bytes
912c7e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
import wandb
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns


def exp_name_from_run(run_config: dict) -> str:
    model = run_config["model"]["_target_"]
    backbone = run_config["model"]["obs_encoder"]["_target_"]
    noise_type = run_config["model"]["noise_type"] if "noise_type" in run_config["model"] else None
    if model == "pfp.policy.fm_so3_policy.FMSO3Policy":
        exp_name = "pfp_so3"
    elif model == "pfp.policy.fm_policy.FMPolicy" and noise_type == "gaussian":
        exp_name = "pfp_euclid"
    elif model == "pfp.policy.fm_policy.FMPolicy" and noise_type == "igso3":
        exp_name = "pfp_euclid_igso3"
    elif (
        model == "pfp.policy.ddim_policy.DDIMPolicy"
        and backbone == "pfp.backbones.pointnet.PointNetBackbone"
    ):
        exp_name = "pfp_ddim"
    elif model == "pfp.policy.fm_so3_policy.FMSO3PolicyImage":
        exp_name = "pfp_images"
    elif backbone == "pfp.backbones.mlp_3dp.MLP3DP":
        exp_name = "dp3"
    elif model == "pfp.policy.fm_policy.FMPolicyImage":
        exp_name = "adaflow"
    elif model == "pfp.policy.ddim_policy.DDIMPolicyImage":
        exp_name = "diffusion_policy"
    else:
        exp_name = "other"
        # raise ValueError(f"Unknown experiment name from model: {model} and backbone: {backbone}")

    # Tunings
    if run_config["model"].get("noise_type") == "biased":
        exp_name += "biased"
    if run_config["model"].get("snr_sampler") == "logit_normal":
        exp_name += "_logitnorm"
    return exp_name


pd.set_option("display.precision", 2)
api = wandb.Api()
runs = api.runs("rl-lab-chisari/pfp-eval-rebuttal")

data_list = []
for run in runs:
    if run.state in ["running", "failed", "crashed"]:
        continue
    exp_name = exp_name_from_run(run.config)
    if exp_name in ["other", "pfp_images", "dp3", "adaflow", "diffusion_policy"]:
        continue
    assert run.summary["episode"] == 99, "Not all runs have 100 episodes"
    data = {
        "task_name": run.config["env_runner"]["env_config"]["task_name"],
        "exp_name": exp_name,
        "k_steps": run.config["policy"]["num_k_infer"],
        "success_rate": run.summary["success"]["mean"] * 100,
    }
    data_list.append(data)


rows = list(
    [
        "pfp_ddim",
        "pfp_so3",
    ]
)
columns = [
    "unplug_charger",
    "close_door",
    "open_box",
    "open_fridge",
    "take_frame_off_hanger",
    "open_oven",
    "put_books_on_bookshelf",
    "take_shoes_out_of_box",
]
data_frame = pd.DataFrame.from_records(data_list)
comparison_frame = data_frame.groupby(["task_name", "exp_name", "k_steps"])
exp_count = comparison_frame.size().unstack(level=0)
exp_count = exp_count.reindex(columns=columns)
# print exp_count with yellow color for cells with other than 3 runs
exp_count = exp_count.style.applymap(lambda x: "background-color: yellow" if x != 3 else "")
# Add more space between rows and columns
paddings = [
    ("padding-right", "20px"),
    ("padding-left", "20px"),
    ("padding-bottom", "10px"),
    ("padding-top", "10px"),
]
exp_count.set_table_styles(
    [
        {
            "selector": "th, td",
            "props": paddings,
        }
    ]
)

# Add horizontal line only after each k_step==16
slice_ = pd.IndexSlice[pd.IndexSlice[:, 16], :]
exp_count.set_properties(**{"border-bottom": "1px solid black"}, subset=slice_)

# Set number precision
exp_count.format("{:.0f}")
exp_count.to_html("experiments/ablation_count.html")

# Process exp_mean DataFrame
exp_mean = comparison_frame.mean()["success_rate"].unstack(level=0)
exp_mean = exp_mean.reindex(columns=columns)

# add a column with the mean of all columns
exp_mean["Mean"] = exp_mean.mean(axis=1)


# Apply green color for cells with the highest value in each column
def highlight_max(s):
    return ["background-color: lightgreen" if v == s.max() else "" for v in s]


# exp_mean_styled = exp_mean.style.apply(highlight_max, axis=0)
exp_mean_styled = exp_mean.style.apply(highlight_max, axis=0)

# Add more space between rows and columns
exp_mean_styled = exp_mean_styled.set_table_styles([{"selector": "th, td", "props": paddings}])

# Add horizontal line only after each K-steps==16
slice_ = pd.IndexSlice[pd.IndexSlice[:, 16], :]
exp_mean_styled.set_properties(**{"border-bottom": "1px solid black"}, subset=slice_)

# Set number precision
exp_mean_styled = exp_mean_styled.format("{:.1f}")

# Save exp_mean to HTML
exp_mean_styled.to_html("experiments/ablation_mean.html")


# ####### Make line plot ###########
ax = sns.relplot(
    data=data_frame[data_frame["exp_name"].isin(["pfp_euclid", "pfp_ddim"])],
    kind="line",
    x="k_steps",
    y="success_rate",
    hue="exp_name",
    hue_order=["pfp_euclid", "pfp_ddim"],
    errorbar=None,
    marker="o",
    markersize=8,
    legend=False,
    aspect=1.5,
)
ax.set_xlabels("K Inference Steps")
ax.set_ylabels("Success Rate")
plt.xscale("log")
plt.minorticks_off()
plt.xticks([1, 2, 4, 8, 16], [1, 2, 4, 8, 16])
plt.legend(
    title="",
    labels=["CFM", "DDIM"],
    # bbox_to_anchor=(0.15, 0.8, 0.8, 0.8),
    # loc="lower right",
    # mode="expand",
    # borderaxespad=0.0,
)

plt.savefig("experiments/ablation_plot.png")

print("Done")